Non-intrusive load monitoring method, medium and electronic equipment

Abstract

The invention discloses a non-intrusive load monitoring method based on a standard deviation composite sliding window and image feature binarization, and the method comprises the steps: carrying out the filtering processing of voltage and current data through employing a three-point median filtering algorithm, constructing an improved standard deviation composite sliding window, and carrying out the two-side detection of a load state change. A standard deviation window is introduced to improve the detection precision of a low-current electric appliance, and whether an event occurs or not is judged through sliding accumulation. The method comprises the following steps: carrying out difference on detected electrical data such as current before and after an event to obtain a new data set, drawing a V-I track image, and extracting V-I track area characteristics at a load state change moment by using image binaryzation; and combining the V-I track characteristics with seasonal electrical appliance probability characteristics, and optimizing convolutional neural network parameters by adopting a GEP algorithm to obtain a GEP-CNN load recognition algorithm so as to accurately decompose and recognize the states of various electrical devices.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grid, and in particular relates to a non-intrusive load monitoring method based on improved standard deviation composite sliding window and image feature binarization. Background Art

[0002] With the rapid advancement of smart grid technology, in-depth exploration of "smart electricity use" and "green electricity use" strategies has become a trend of the times. As the core driving force for the modernization of the national power system, smart grids not only provide users with the ability to monitor the energy efficiency of electrical appliances in real time, but also its continuous technological innovation and infrastructure construction herald the improvement of power resource utilization and the quality of life of residents. As the blueprint for the future power system, smart grids integrate hardware, high-precision sensing and measurement technology, high-speed communication technology, and advanced control systems, aiming to build a safe, reliable, economical, efficient and environmentally friendly power grid ecosystem.

[0003] Although power system load monitoring is an indispensable function of smart grids, traditional intrusive monitoring methods require the deployment of monitoring equipment for each load separately, which not only leads to a sharp increase in hardware costs (including procurement, installation and maintenance), but also because each monitoring point is an independent data island. When faced with high-load situations, data processing and analysis become extremely complex and inefficient.

[0004] In this context, how to apply non-intrusive load decomposition technology in smart grids has become the current mainstream trend. According to the concept, this technology only needs to install a single monitoring device at the user's incoming line end, and deeply mine and analyze the collected data through advanced algorithms to accurately identify and analyze the composition and working status of various loads in the system. This breakthrough can not only greatly reduce the deployment cost and complexity of monitoring equipment, but also realize all-weather and refined monitoring of various loads. For users, non-intrusive load decomposition technology is like an intelligent housekeeper, which can not only immediately warn of potential abnormalities or failures, but also help users understand the details of electricity consumption and promote the improvement and practice of energy conservation and emission reduction awareness. From a national strategic perspective, the promotion of this technology has far-reaching and significant significance for accelerating the green transformation of my country's energy structure and promoting the realization of energy conservation and emission reduction goals. Summary of the invention

[0005] The present invention aims to solve one of the technical problems existing in the related art at least to a certain extent.

[0006] One object of the present invention is to provide a non-invasive load monitoring method based on improved standard deviation composite sliding window and image feature binarization, so as to realize power system load identification, enable the load to be reasonably identified and monitored, improve data processing efficiency, and ensure the safe and stable operation of the power system.

[0007] In order to achieve the above-mentioned object, the present invention provides a non-intrusive load monitoring method, comprising:

[0008] Collecting electrical signal data of sample electrical equipment and preprocessing it to obtain a data set;

[0009] Conduct seasonal cluster analysis on electrical appliances according to their usage frequency, and establish a seasonal electrical appliance probability feature table;

[0010] Set the standard deviation composite sliding window and define the parameters of the standard deviation composite sliding window;

[0011] The sliding window mechanism is used to detect the change event of the electrical appliance state, and the electrical data before and after the change event are differentially processed to obtain the difference data set;

[0012] Draw the voltage-current trajectory diagram based on the difference data set, extract the trajectory features, and establish the load switching state feature library;

[0013] The optimization algorithm is used to optimize the parameters of the deep learning model, and then the optimized deep learning model is trained according to the data set and the established load switching state feature library to obtain the load identification model;

[0014] The load identification model is used to identify the electrical appliances in the electrical appliance data set to be tested, and the seasonal electrical appliance probability feature table is combined for mutual correction.

[0015] A further preferred technical solution of the present invention is that the electrical signal data of the sample electrical equipment is collected and preprocessed to obtain a data set; including:

[0016] Collect electrical signal data of sample electrical equipment, perform outlier processing on the data, and remove abnormal data values;

[0017] The electrical data with outliers removed is filtered using a three-point median filtering algorithm, and the electrical data is smoothed and fitted to remove noise and retain the high-frequency components and trend information of the signal, thereby generating a data set.

[0018] Preferably, the step of setting a standard deviation composite sliding window and defining parameters of the standard deviation composite sliding window comprises:

[0019] Set the standard deviation composite sliding window to detect the state change of electrical equipment; the specific steps are:

[0020] Two load windows are set, namely, the mean calculation window Wm and the transient detection window Wz, wherein the transient detection window Wz consists of a transient detection part Wzd and a standard deviation detection part Wzb, the transient detection part Wzd corresponds to a window length zd, the standard deviation detection part Wzb corresponds to a window length zb, and the window length z of the transient detection window Wz is z=zd+zb;

[0021] Define the parameters of the standard deviation composite sliding window, including mean and standard deviation, for event detection; the specific steps are:

[0022] Define the mean value of transient detection window Wz as M z , transient detection part Wzd mean M zd , the mean value M of the standard deviation detection part Wzb zb , and the standard deviation value B of the standard deviation detection part Wzb;

[0023] Defining the forward cumulative sum and negative cumulative sum Used to detect event input and output;

[0024] Define the length of each sliding window, the threshold for cumulative event occurrence, and the standard deviation threshold parameters.

[0025] As a preference, the mean value of the transient detection window Wz is M z , transient detection part Wzd mean M zd , the mean value M of the standard deviation detection part Wzb zb , and the calculation formula of the standard deviation value B of the standard deviation detection part Wzb are respectively expressed as:

[0026]

[0027] Among them, k is the leftmost data bit of the current sliding window, m is the length of the mean calculation window, and i(j) is the current value data of the jth data point.

[0028] Positive Cumulative Sum and negative cumulative sum The calculation formulas are expressed as:

[0029]

[0030] l=1+ad

[0031] Where λ is the proportional coefficient, α is the weight coefficient, d is the time factor, Y is the standard deviation threshold, and β is the maximum value of the fluctuation when no event occurs.

[0032] Preferably, the method of using a sliding window mechanism to detect a change event of an electrical appliance state and performing differential processing on electrical data before and after the change event to obtain a differential data set comprises:

[0033] The data set is brought into the standard deviation composite sliding window of the defined parameters for sliding, and the cumulative sum is calculated to determine whether the cumulative sum is greater than the threshold;

[0034] If it is not greater than the threshold, it continues to slide; if it is greater than the threshold, it indicates that a switching event has occurred, which is recorded as the time factor d;

[0035] Continue sliding the window to the positive cumulative sum Less than the threshold and g k + £g k-1 + , is the positive cumulative sum of the previous data point, and B must be lower than the preset threshold to be considered as entering a steady state, and the time point t is defined s , the time factor parameter d is cleared, and the sliding window continues until all the data slides through the standard deviation composite sliding window;

[0036] Negative cumulative sum Processing method and positive cumulative sum same;

[0037] The current and voltage before and after the load state change detected by the event d are subtracted to obtain a voltage and current difference data set.

[0038] Preferably, the method of drawing a voltage-current trajectory diagram based on the difference data set, extracting trajectory features, and establishing a load switching state feature library comprises:

[0039] Draw a voltage-current trajectory diagram based on the difference data set, perform image binarization on the obtained voltage-current trajectory diagram, process the enclosed area as white, and process the rest as black;

[0040] After normalizing the voltage and current, the proportion of the white part is obtained. Multiplying it by the normalization factor can obtain the voltage-current trajectory area characteristics, thereby establishing a load switching state feature library.

[0041] Preferably, the optimization algorithm is used to optimize the deep learning model parameters, and then the optimized deep learning model is trained according to the data set and the established load switching state feature library to obtain the load identification model; including:

[0042] Select the Gene Expression Programming algorithm as the optimization algorithm for the deep learning model, set the population size and genetic algebra of the Gene Expression Programming algorithm, use the roulette wheel method for selection, single-point crossover for crossover, and uniform mutation for mutation;

[0043] Use the gene expression programming algorithm to optimize the parameters of convolutional neural networks;

[0044] The data sets before and after the load state change d time detected by the event and the load switching state feature library are brought into the optimized convolutional neural network, and the convolutional neural network is trained to obtain the GEP-CNN load identification algorithm model.

[0045] Preferably, the method of using the load identification model to identify the electrical appliance data set to be detected, and performing mutual correction in combination with the seasonal electrical appliance probability feature table, comprises:

[0046] The appliance data set to be tested is brought into the trained load identification model, and the identification results are corrected with the seasonal appliance probability characteristic table. If the appliances identified by the load identification model are significantly different from those in the seasonal appliance probability characteristic table, the test data set is re-input into the optimized decomposition algorithm and multiple decomposition and identification are performed. If the results are consistent, the output is completed and the appliance is added to the seasonal appliance probability characteristic table to complete the load identification.

[0047] Another aspect of the present invention provides a non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions enabling a computer to execute the above-mentioned non-intrusive load monitoring method.

[0048] On the other hand, the present invention provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls the logic instructions in the memory to execute the above-mentioned non-intrusive load monitoring method.

[0049] Beneficial effects: (1) Compared with the cumbersome configuration process of the invasive method, the non-invasive load identification of the present invention only requires the deployment of monitoring equipment at the total collection end of the user entrance. Through the precise measurement of the collected data, it is possible to accurately achieve a comprehensive and high-precision control of the power consumption. For power grid companies and users, non-invasive load decomposition is like a diligent and responsible "power steward", always alert to possible abnormal faults, and ensure safe and worry-free power use. At the same time, it also helps users to deeply understand their own power consumption habits, helps users to accurately grasp energy consumption, and efficiently implement energy conservation and management strategies.

[0050] (2) The present invention proposes a non-invasive load monitoring method based on an improved standard deviation composite sliding window and image feature binarization. Compared with ordinary sliding windows, the present invention introduces a standard deviation sliding window to improve the accuracy of event detection and the time it takes for the load to enter a steady state. Feature extraction based on image feature binarization and the use of image proportion to calculate area features are simple and accurate. The GEP algorithm is also used to optimize the parameters of the convolutional neural network to the GEP-CNN load identification algorithm. Combined with the probability of users using electrical appliances in different seasons, it can achieve effective identification of the status of various household appliances in different scenarios. Compared with the invasive method, the monitoring device is only installed at the main end of the power system, which saves a lot of monitoring equipment, reduces data processing and analysis work, and at the same time realizes the status monitoring of each type of load in the system in disguise.

[0051] (3) The present invention performs outlier detection on the collected data to remove outliers, and performs preprocessing using a three-point median filter algorithm. The data set is input into the improved standard deviation composite sliding window, and the start / stop state change of the appliance is determined by judging whether the sliding cumulative sum is greater than the threshold of the appliance state change. The standard deviation window is introduced to determine the degree of data fluctuation. When the window standard deviation is less than the threshold, it is considered a low-power load switching event, increasing the cumulative sum of the sliding window state change, thereby effectively identifying the start / stop events of low-current appliances.

[0052] (4) The present invention constructs a seasonal electrical appliance probability feature table through a clustering algorithm based on the user's habits of using electrical appliances in different seasons. The current and voltage before and after the load state change after event detection are subtracted to obtain a new data set. The VI trajectory diagram is drawn based on the difference, and the closed image is binarized and filled. The area surrounded by the curve is processed as white, and the other parts are processed as black. The voltage and current are normalized to obtain the proportion of the white part, and then multiplied by the normalization multiple to obtain the VI trajectory area characteristics of the load state change after event detection. The seasonal electrical appliance probability feature table and the VI trajectory area characteristics are combined for load identification.

[0053] (4) The present invention introduces the GEP algorithm into the CNN model to optimize the parameters of the convolutional neural network model. The optimal solution is gradually searched and the optimal parameters are selected as the parameters of the CNN algorithm. The GEP-CNN model is trained with the training set, and the test set is input into the trained GEP-CNN model. Finally, the various electrical load states are combined with the probability distribution diagram of each seasonal electrical appliance to obtain the final load decomposition. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The figure is a flow chart of the non-intrusive load monitoring method of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0056] In view of the current characteristics of a wide variety of power equipment and diverse user characteristics, the present invention proposes a non-invasive load monitoring method based on standard deviation composite sliding window and image feature binarization. The collected load data is first preprocessed, and an improved load sliding window bilateral is constructed to detect load state changes. The standard deviation window is introduced to improve the detection accuracy of small current electrical appliances. The image features of the time period when the event occurs are binarized, and the trajectory features of the load state change are extracted. The seasonal use of electrical appliances is integrated to construct a seasonal load feature table. Seasonal load features are used as new load features to assist load decomposition. The multi-feature GEP algorithm is used to optimize the convolutional neural network parameters to obtain the GEP-CNN load recognition algorithm, which combines VI image features with seasonal electrical appliance probability characteristics to accurately decompose and identify the status of various electrical equipment. The present invention is based on the non-invasive load monitoring technology of standard deviation composite sliding window and image feature binarization. The main considerations are three aspects: (1) how to use the improved load sliding window bilaterally to accurately detect the load state changes of small current electrical appliances; (2) how to extract and establish a feature library based on the seasonal probability characteristics and VI characteristics of users; (3) how to use genetic expression programming (GEP) to optimize the parameters of the load identification algorithm.

[0057] The present invention provides a non-intrusive load monitoring method, medium and electronic device, which are described in detail through the following embodiments.

[0058] First, the abbreviations of technical terms that may be involved in the following embodiments are explained:

[0059] VI Trajectory Graph: Voltage-Current Trajectory Graph is a graph used to represent the relationship between voltage (V) and current (I) of an appliance under different working conditions. This graph is commonly used in the field of power systems and electrical engineering, especially in non-intrusive load monitoring (NILM) technology, to analyze and identify the energy consumption mode and operating status of appliances.

[0060] CNN: Convolutional Neural Network, a deep learning model widely used in image and video recognition, classification, and segmentation.

[0061] GEP: Gene Expression Programming (GEP) is an adaptive evolutionary algorithm developed from Genetic Algorithms (GA) and Genetic Programming (GP). The core idea of ​​GEP is to solve complex problems by simulating the structure and function of biological genes using simple coding methods. It absorbs the advantages of GA and GP while overcoming their shortcomings, showing a significant improvement in processing noisy data and generalization capabilities.

[0062] Embodiment 1: This embodiment provides a non-intrusive load monitoring method. Figure 1 As shown, including:

[0063] S1. Collect electrical signal data of sample electrical equipment, perform outlier processing on the data, remove abnormal data values, and remove the impact of abnormal data on subsequent event detection, feature extraction, and load identification.

[0064] S2. Use a three-point median filtering algorithm to filter the electrical data with outliers removed, smooth and fit the electrical data, remove noise, and retain the high-frequency components and trend information of the signal, thereby generating a data set.

[0065] S3. Perform cluster analysis on each appliance according to the frequency of users using the equipment in different seasons. Divide the year into four seasons, conduct probability statistics on the frequency of users using different appliances in each season, and make a seasonal appliance probability feature table.

[0066] Clustering is divided into four seasons: spring: March-May, summer: June-August, autumn: September-November, and winter: December-February. The usage probability characteristics of electrical appliances in different seasons are obtained based on the clustering algorithm.

[0067] S4. Set two load windows, namely the mean calculation window Wm and the transient detection window Wz, wherein the transient detection window Wz consists of a transient detection part Wzd and a standard deviation detection part Wzb, the transient detection part Wzd corresponds to a window length zd, the standard deviation detection part Wzb corresponds to a window length zb, and the window length z of the transient detection window Wz is z=zd+zb.

[0068] S5. Define the mean value of transient detection window Wz as M z, transient detection part Wzd mean M zd , the mean value M of the standard deviation detection part Wzb zb , and the standard deviation value B of the standard deviation detection part Wzb; the calculation formulas are respectively expressed as:

[0069]

[0070]

[0071] Among them, k is the leftmost data bit of the current sliding window, m is the length of the mean calculation window, and i(j) is the current value data of the jth data point.

[0072] S6. Define the positive cumulative sum and negative cumulative sum Used to detect event input and output, define parameters such as the length of each sliding window, the threshold for cumulative event occurrence, and the standard deviation threshold; positive cumulative sum and negative cumulative sum The calculation formulas are expressed as:

[0073]

[0074] l=1+ad

[0075] Where λ is the proportional coefficient, α is the weight coefficient, d is the time factor, Y is the standard deviation threshold, and β is the maximum value of the fluctuation when no event occurs.

[0076] S7, bringing the data set processed in step S2 into the standard deviation composite sliding window of defined parameters for sliding, with a sliding step of 1, and calculating the cumulative sum to determine whether the cumulative sum is greater than a threshold;

[0077] S8, continue sliding the window to the positive cumulative sum Less than the threshold and g k + £g k-1 + , is the positive cumulative sum of the previous data point, and B must be lower than the preset threshold to be considered as entering a steady state, and the time point t is defined s , the time factor parameter d is cleared, and the sliding window continues until all the data slides through the standard deviation composite sliding window;

[0078] Negative cumulative sum Processing method and positive cumulative sum same;

[0079] S9. Differences are taken between the current and voltage before and after the load state change detected by the event to obtain a voltage and current difference data set.

[0080] S10, draw a VI trajectory diagram based on the difference, binarize the image, treat the enclosed area as white, and treat the other parts as black. After normalizing the voltage and current, the proportion of the white part is obtained, and multiplied by the normalization multiple to obtain the VI trajectory area feature and establish a load switching state feature library.

[0081] The transient voltages before and after the switching event are averaged, and the current is subtracted to form a circular data structure. The data points of U and I are plotted. After the VI trajectory is saved as a picture, the VI picture is binarized and the area feature of the VI trajectory is calculated.

[0082] S11. Set the population size and genetic generation of GEP, use roulette wheel for selection, single-point crossover for crossover, and uniform mutation for mutation;

[0083] S12. Use the GEP algorithm to optimize the parameters of the convolutional neural network, and then train the optimized deep learning model based on the data set before and after the load state changes d times detected by the event and the established load switching state feature library to obtain the GEP-CNN load identification algorithm model.

[0084] S13, bring the data set to be tested into the GEP-CNN load identification algorithm model, and correct it with the seasonal probability feature table of electrical appliances. If the electrical appliances of the identification algorithm are significantly different from those in the seasonal probability feature table, re-input the test data set into the optimized decomposition algorithm, perform multiple decomposition and identification, and if the results are consistent, complete the output and add the electrical appliances to the seasonal probability feature table to complete the load identification.

[0085] In step S12 and step S13, the data set is divided into 70% training set and 30% test set. The training set is used to train the GEP-CNN load identification algorithm model, and the test set is input into the trained network to output the decomposition results of various electrical load states.

[0086] Embodiment 2: This embodiment provides a non-transitory computer-readable storage medium on which computer instructions are stored. The computer instructions enable a computer to execute a non-intrusive load monitoring method in Embodiment 1.

[0087] Embodiment 3: This embodiment provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. The processor may call the logic instructions in the memory to execute a non-intrusive load monitoring method in Embodiment 1.

[0088] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0089] Embodiment 4: This embodiment provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a non-intrusive load monitoring method in Embodiment 1.

[0090] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0091] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A non-intrusive load monitoring method, characterized in that: include: Collecting electrical signal data of sample electrical equipment and preprocessing it to obtain a data set; Conduct seasonal cluster analysis on electrical appliances according to their usage frequency, and establish a seasonal electrical appliance probability feature table; Set the standard deviation composite sliding window and define the parameters of the standard deviation composite sliding window; The sliding window mechanism is used to detect the change event of the electrical appliance state, and the electrical data before and after the change event are differentially processed to obtain the difference data set; Draw the voltage-current trajectory diagram based on the difference data set, extract the trajectory features, and establish the load switching state feature library; The optimization algorithm is used to optimize the parameters of the deep learning model, and then the optimized deep learning model is trained according to the data set and the established load switching state feature library to obtain the load identification model; The load identification model is used to identify the electrical appliances in the electrical appliance data set to be tested, and the seasonal electrical appliance probability feature table is combined for mutual correction.

2. The non-intrusive load monitoring method according to claim 1, characterized in that: The method of collecting electrical signal data of sample electrical equipment and performing preprocessing to obtain a data set includes: Collect electrical signal data of sample electrical equipment, perform outlier processing on the data, and remove abnormal data values; The electrical data with outliers removed are filtered using a three-point median filtering algorithm, and the electrical data are smoothed and fitted to remove noise while retaining the high-frequency components and trend information of the signal, thereby generating a data set.

3. The non-intrusive load monitoring method according to claim 1, characterized in that: The step of setting a standard deviation composite sliding window and defining parameters of the standard deviation composite sliding window includes: Set the standard deviation composite sliding window to detect the state change of electrical equipment; the specific steps are: Two load windows are set, namely, the mean calculation window Wm and the transient detection window Wz, wherein the transient detection window Wz consists of a transient detection part Wzd and a standard deviation detection part Wzb, the transient detection part Wzd corresponds to a window length zd, the standard deviation detection part Wzb corresponds to a window length zb, and the window length z of the transient detection window Wz is z=zd+zb; Define the parameters of the standard deviation composite sliding window, including mean and standard deviation, for event detection; the specific steps are: Define the mean value of transient detection window Wz as M z , transient detection part Wzd mean M zd , the mean value M of the standard deviation detection part Wzb zb , and the standard deviation value B of the standard deviation detection part Wzb; Defining the forward cumulative sum and negative cumulative sum Used to detect event input and output; Define the length of each sliding window, the threshold for cumulative event occurrence, and the standard deviation threshold parameters.

4. The non-intrusive load monitoring method according to claim 3, characterized in that: The mean value of transient detection window Wz is M z , transient detection part Wzd mean M zd , the mean value M of the standard deviation detection part Wzb zb , and the calculation formula of the standard deviation value B of the standard deviation detection part Wzb are respectively expressed as: Among them, k is the leftmost data bit of the current sliding window, m is the length of the mean calculation window, and i(j) is the current value data of the jth data point. Positive Cumulative Sum and negative cumulative sum The calculation formulas are expressed as: l=1+ad Where λ is the proportional coefficient, α is the weight coefficient, d is the time factor, Y is the standard deviation threshold, and β is the maximum value of the fluctuation when no event occurs.

5. The non-intrusive load monitoring method according to claim 4, characterized in that: The method uses a sliding window mechanism to detect the change event of the electrical appliance state, and performs differential processing on the electrical data before and after the change event to obtain a differential data set; including: The data set is brought into the standard deviation composite sliding window of the defined parameters for sliding, and the cumulative sum is calculated to determine whether the cumulative sum is greater than the threshold; If it is not greater than the threshold, it continues to slide; if it is greater than the threshold, it indicates that a switching event has occurred, which is recorded as the time factor d; Continue sliding the window to the positive cumulative sum Less than the threshold and is the positive cumulative sum of the previous data point, and B must be lower than the preset threshold to be considered as entering a steady state, and the time point t is defined s , the time factor parameter d is cleared, and the sliding window continues until all the data slides through the standard deviation composite sliding window; Negative cumulative sum Processing method and positive cumulative sum same; The current and voltage before and after the load state change detected by the event d are subtracted to obtain a voltage and current difference data set.

6. The non-intrusive load monitoring method according to claim 1, characterized in that: The method of drawing a voltage-current trajectory diagram based on the difference data set, extracting trajectory features, and establishing a load switching state feature library comprises: Draw a voltage-current trajectory diagram based on the difference data set, perform image binarization on the obtained voltage-current trajectory diagram, process the enclosed area as white, and process the rest as black; After normalizing the voltage and current, the proportion of the white part is obtained. Multiplying it by the normalization factor can obtain the voltage-current trajectory area characteristics, thereby establishing a load switching state feature library.

7. The non-intrusive load monitoring method according to claim 1, characterized in that: The method uses an optimization algorithm to optimize the parameters of the deep learning model, and then trains the optimized deep learning model according to the data set and the established load switching state feature library to obtain the load identification model; including: Select the Gene Expression Programming algorithm as the optimization algorithm for the deep learning model, set the population size and genetic algebra of the Gene Expression Programming algorithm, use the roulette wheel method for selection, single-point crossover for crossover, and uniform mutation for mutation; Use the gene expression programming algorithm to optimize the parameters of convolutional neural networks; The data set and load switching state feature library are brought into the optimized convolutional neural network, the convolutional neural network is trained, and the GEP-CNN load identification algorithm model is obtained.

8. The non-intrusive load monitoring method according to claim 1, characterized in that: The method of using the load identification model to identify the electrical appliance data set to be detected and performing mutual correction in combination with the seasonal electrical appliance probability feature table comprises: The appliance data set to be tested is brought into the trained load identification model, and the identification results are corrected with the seasonal appliance probability characteristic table. If the appliances identified by the load identification model are significantly different from those in the seasonal appliance probability characteristic table, the test data set is re-input into the optimized decomposition algorithm and multiple decomposition and identification are performed. If the results are consistent, the output is completed and the appliance is added to the seasonal appliance probability characteristic table to complete the load identification.

9. A non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions enabling a computer to execute the non-intrusive load monitoring method according to any one of claims 1 to 8.

10. An electronic device comprising: A processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls the logic instructions in the memory to execute the non-intrusive load monitoring method described in any one of claims 1 to 8.

Images